pytorch

Calling super's forward() method

由 与世无争的帅哥 提交于 2020-12-30 05:44:49
问题 What is the most appropriate way to call the forward() method of a parent Module ? For example, if I subclass the nn.Linear module, I might do the following class LinearWithOtherStuff(nn.Linear): def forward(self, x): y = super(Linear, self).forward(x) z = do_other_stuff(y) return z However, the docs say not to call the forward() method directly: Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since

Calling super's forward() method

由 孤街醉人 提交于 2020-12-30 05:43:59
问题 What is the most appropriate way to call the forward() method of a parent Module ? For example, if I subclass the nn.Linear module, I might do the following class LinearWithOtherStuff(nn.Linear): def forward(self, x): y = super(Linear, self).forward(x) z = do_other_stuff(y) return z However, the docs say not to call the forward() method directly: Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since

pytorch, AttributeError: module 'torch' has no attribute 'Tensor'

由 半世苍凉 提交于 2020-12-29 08:59:26
问题 I'm working with Python 3.5.1 on a computer having CentOS Linux 7.3.1611 (Core) operating system. I'm trying to use PyTorch and I'm getting started with this tutorial. Unfortunately, the #4 line of the example creates troubles: >>> torch.Tensor(5, 3) Traceback (most recent call last): File "<stdin>", line 1, in <module> AttributeError: module 'torch' has no attribute 'Tensor' I cannot understand this error... of course in Torch the 'torch' does have an attribute 'Tensor'. The same command

深入理解卷积网络的卷积

由 谁都会走 提交于 2020-12-28 10:07:00
卷积神经网络是一种特殊的神经网络结构，是自动驾驶汽车、人脸识别系统等计算机视觉应用的基础，其中基本的矩阵乘法运算被卷积运算取代。它们专门处理具有网格状拓扑结构的数据。例如，时间序列数据和图像数据可以看作是一个二维像素网格。 历史 卷积神经网络最初是由福岛核电站在1980年引入的，当时名为Neocognitron。它的灵感来自于Hubel和Weisel提出的神经系统的层次模型。但由于其复杂的无监督学习算法，即无监督学习，该模型并不受欢迎。1989年，Yann LeCun利用反向传播和Neocognitron的概念提出了一种名为LeNet的架构，该架构被美国和欧洲用于手写的邮政编码识别。邮政服务。Yann LeCun进一步研究了这个项目，最终在1998年发布了LeNet-5——第一个引入了我们今天在CNN仍然使用的一些基本概念的现代卷积神经网络。他还发布了MNIST手写数字数据集，这可能是机器学习中最著名的基准数据集。在20世纪90年代，计算机视觉领域转移了它的焦点，许多研究人员停止了对CNN架构的研究。神经网络的研究经历了一个寒冷的冬天，直到2012年，多伦多大学的一组研究人员在著名的ImageNet挑战赛中进入了一个基于CNN的模型(AlexNet)，最终以16.4%的错误率赢得了比赛。此后，卷积神经网络不断向前发展，基于CNN的体系结构不断赢得ImageNet, 2015年

How to use multiple GPUs in pytorch?

由 浪子不回头ぞ 提交于 2020-12-28 07:00:57
问题 I am learning pytorch and follow this tutorial: https://pytorch.org/tutorials/beginner/transfer_learning_tutorial.html I use this command to use a GPU. device = torch.device("**cuda:0**" if torch.cuda.is_available() else "cpu") But, I want to use two GPUs in jupyter , like this: device = torch.device("**cuda:0,1**" if torch.cuda.is_available() else "cpu") Of course, this is wrong. So, How can I do this? 回答1: Using multi-GPUs is as simply as wrapping a model in DataParallel and increasing the

Saving and reload huggingface fine-tuned transformer

由 我们两清 提交于 2020-12-26 11:11:18
问题 I am trying to reload a fine-tuned DistilBertForTokenClassification model. I am using transformers 3.4.0 and pytorch version 1.6.0+cu101. After using the Trainer to train the downloaded model, I save the model with trainer.save_model() and in my trouble shooting I save in a different directory via model.save_pretrained(). I am using Google Colab and saving the model to my Google drive. After testing the model I also evaluated the model on my test getting great results, however, when I return

Saving and reload huggingface fine-tuned transformer

由 一世执手 提交于 2020-12-26 11:10:03
问题 I am trying to reload a fine-tuned DistilBertForTokenClassification model. I am using transformers 3.4.0 and pytorch version 1.6.0+cu101. After using the Trainer to train the downloaded model, I save the model with trainer.save_model() and in my trouble shooting I save in a different directory via model.save_pretrained(). I am using Google Colab and saving the model to my Google drive. After testing the model I also evaluated the model on my test getting great results, however, when I return

How can I chunk a PyTorch tensor into a specified bucket size with overlap?

由 点点圈 提交于 2020-12-26 11:07:43
问题 Specifically, I have a tensor of shape: torch.Size([1, 16]) I want to bucket this into 7 buckets (of 4 each). Example: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16] should become: [[1, 2, 3, 4], [3, 4, 5, 6], [5, 6, 7, 8], [7, 8, 9, 10], [9, 10, 11, 12], [11, 12, 13, 14], [13, 14, 15, 16], ] How can I achieve this with PyTorch? 回答1: Looks like an unfold: t.unfold(0,4,2) Output: tensor([[ 1., 2., 3., 4.], [ 3., 4., 5., 6.], [ 5., 6., 7., 8.], [ 7., 8., 9., 10.], [ 9., 10., 11., 12.],

How can I chunk a PyTorch tensor into a specified bucket size with overlap?

由 扶醉桌前 提交于 2020-12-26 11:06:04
问题 Specifically, I have a tensor of shape: torch.Size([1, 16]) I want to bucket this into 7 buckets (of 4 each). Example: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16] should become: [[1, 2, 3, 4], [3, 4, 5, 6], [5, 6, 7, 8], [7, 8, 9, 10], [9, 10, 11, 12], [11, 12, 13, 14], [13, 14, 15, 16], ] How can I achieve this with PyTorch? 回答1: Looks like an unfold: t.unfold(0,4,2) Output: tensor([[ 1., 2., 3., 4.], [ 3., 4., 5., 6.], [ 5., 6., 7., 8.], [ 7., 8., 9., 10.], [ 9., 10., 11., 12.],

在数据科学方面，python和R有何区别?

由 前提是你 提交于 2020-12-25 09:11:14
　　python和R都是当下比较流行的编程语言，拥有强大的生态系统和社区，受到大家的追捧和喜欢，那么在数据科学方面，python和R有何区别?我们来看看吧。 　　大多数深度学习研究都使用python完成的，因此Keras和PyTorch之类的工具具有python优先的开发，你可以在Keras的深度学习简介中了解这些主题。 　　python和R之上具有优势的另一个领域就是将模型部署到其他软件中，python是一种通用的编程语言，因此，如果您在使用python编写应用程序，包含基于python的模型的过程是无缝的。我们在python设计机器学习工作流中介绍了部署模型和构建python的数据工程管道。 　　python通常被认为是一种通用语言，具有易于理解的语法。 　　在R中进行大流的统计建模研究，有多种模型可供选择。R的另一个大窍门是使用Shiny轻松创建仪表板，对于没有太多技术经验的人来说，可以创建发布仪表板并进行分享。 　　R的功能在开发时考虑了统计学家的问题，从而赋予特定领域的优势，比如说数据可视化功能。 　　python最初是用于软件开发的编程语言，对于具有计算机科学或者软件开发的人员来说可以更容易使用，而且从其他语言过度到python要比R更加简单。 来源： oschina 链接： https://my.oschina.net/u/4408222/blog/4839636
订阅 pytorch